105 lines · plain
1; RUN: llc -mtriple=amdgcn < %s | FileCheck --check-prefixes=GCN,FUNC %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck --check-prefixes=GCN,FUNC %s3; RUN: llc -mtriple=r600 -mcpu=redwood < %s | FileCheck --check-prefixes=EG,FUNC %s4; RUN: llc -mtriple=r600 -mcpu=cayman < %s | FileCheck --check-prefixes=EG,FUNC %s5 6 7@b = internal addrspace(4) constant [1 x i16] [ i16 7 ], align 28 9@float_gv = internal unnamed_addr addrspace(4) constant [5 x float] [float 0.0, float 1.0, float 2.0, float 3.0, float 4.0], align 410 11; FUNC-LABEL: {{^}}float:12; GCN: s_load_dword13 14; EG: VTX_READ_3215; EG: @float_gv16; EG-NOT: MOVA_INT17; EG-NOT: MOV18define amdgpu_kernel void @float(ptr addrspace(1) %out, i32 %index) {19entry:20 %0 = getelementptr inbounds [5 x float], ptr addrspace(4) @float_gv, i32 0, i32 %index21 %1 = load float, ptr addrspace(4) %022 store float %1, ptr addrspace(1) %out23 ret void24}25 26@i32_gv = internal unnamed_addr addrspace(4) constant [5 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4], align 427 28; FUNC-LABEL: {{^}}i32:29 30; GCN: s_load_dword31 32; EG: VTX_READ_3233; EG: @i32_gv34; EG-NOT: MOVA_INT35; EG-NOT: MOV36define amdgpu_kernel void @i32(ptr addrspace(1) %out, i32 %index) {37entry:38 %0 = getelementptr inbounds [5 x i32], ptr addrspace(4) @i32_gv, i32 0, i32 %index39 %1 = load i32, ptr addrspace(4) %040 store i32 %1, ptr addrspace(1) %out41 ret void42}43 44 45%struct.foo = type { float, [5 x i32] }46 47@struct_foo_gv = internal unnamed_addr addrspace(4) constant [1 x %struct.foo] [ %struct.foo { float 16.0, [5 x i32] [i32 0, i32 1, i32 2, i32 3, i32 4] } ]48 49; FUNC-LABEL: {{^}}struct_foo_gv_load:50; GCN: s_load_dword51 52; EG: VTX_READ_3253; EG: @struct_foo_gv54; EG-NOT: MOVA_INT55; EG-NOT: MOV56define amdgpu_kernel void @struct_foo_gv_load(ptr addrspace(1) %out, i32 %index) {57 %gep = getelementptr inbounds [1 x %struct.foo], ptr addrspace(4) @struct_foo_gv, i32 0, i32 0, i32 1, i32 %index58 %load = load i32, ptr addrspace(4) %gep, align 459 store i32 %load, ptr addrspace(1) %out, align 460 ret void61}62 63@array_v1_gv = internal addrspace(4) constant [4 x <1 x i32>] [ <1 x i32> <i32 1>,64 <1 x i32> <i32 2>,65 <1 x i32> <i32 3>,66 <1 x i32> <i32 4> ]67 68; FUNC-LABEL: {{^}}array_v1_gv_load:69; GCN: s_load_dword70 71; EG: VTX_READ_3272; EG: @array_v1_gv73; EG-NOT: MOVA_INT74; EG-NOT: MOV75define amdgpu_kernel void @array_v1_gv_load(ptr addrspace(1) %out, i32 %index) {76 %gep = getelementptr inbounds [4 x <1 x i32>], ptr addrspace(4) @array_v1_gv, i32 0, i32 %index77 %load = load <1 x i32>, ptr addrspace(4) %gep, align 478 store <1 x i32> %load, ptr addrspace(1) %out, align 479 ret void80}81 82; FUNC-LABEL: {{^}}gv_addressing_in_branch:83 84; EG: VTX_READ_3285; EG: @float_gv86; EG-NOT: MOVA_INT87define amdgpu_kernel void @gv_addressing_in_branch(ptr addrspace(1) %out, i32 %index, i32 %a) {88entry:89 %0 = icmp eq i32 0, %a90 br i1 %0, label %if, label %else91 92if:93 %1 = getelementptr inbounds [5 x float], ptr addrspace(4) @float_gv, i32 0, i32 %index94 %2 = load float, ptr addrspace(4) %195 store float %2, ptr addrspace(1) %out96 br label %endif97 98else:99 store float 1.0, ptr addrspace(1) %out100 br label %endif101 102endif:103 ret void104}105